/5 CONNECTED STEPS
Produce reports whose claims trace to data
Generate a report or chart-backed answer using computed facts and auditable relationships.
Download complete workflow .md ↓Use with your AI
Complete workflow download included with library membership. Try the free starters without signup.
What you need
Verified records with units, subject, timestamps and citations; controllable model inputs and rendering.
Why these fit together
Field semantics determine what can be calculated; code settles relationships; rendering binds facts; a separate relational audit catches misleading claims even when numerals are correct.
Follow this order
Name computed quantities from source semantics
Define what each source value actually measures.
Compute claims before asking AI for prose
Compute relationships before narrative generation.
Ground charts and answers on one data artifact
Feed chart and answer from the same typed artifact.
Trace every rendered number to a source fact
Trace final numerals to declared facts.
Stop AI reports inventing relationships
Check comparisons, scope and subject in final output.
Give this task to your AI
Use the Produce reports whose claims trace to data workflow. My task and inputs are supplied below. Read the prerequisites and check your available tools. Follow the core steps in order, passing the actual output of each step to the next. Adapt installation and commands to this environment; do not copy another agent's configuration blindly. If a required tool or input is missing, identify it before that step. Run the completion check and report observed results and unresolved work.
Setup to provide or build
A shared typed fact schema, deterministic computations, renderer, claim validator and fixtures containing unsupported claims.
Check the result
A deliberately wrong trend and unsupported number are rejected; the chart and answer use the same dated records and citations.
Curated implementation workflow. End-to-end execution must be verified in your environment.
Optional additions
Load these only when your task needs them.